The Fisher effect in the EU revisited: new evidence using panel cointegration estimation with global stochastic trends

Abstract

This article is aimed at verifying the fulfilment of the Fisher hypothesis for a panel of 15 EU countries using the recent developments in the estimation of panel cointegration models with cross-sectional dependence generated by unobservable global stochastic trends (Bai et al., 2009). Bai et al. (2009) propose two iterative procedures that jointly estimate the slope parameters and the stochastic trends. The resulting estimators are referred to, respectively, as Continuously Updated and Bias Corrected (CupBC) and Continuously Updated and Fully Modified (CupFM) estimators. Our results show that, if we ignore the cross-sectional dependence generated by global stochastic trends when estimating the Fisher equation, we could erroneously infer that there is a full Fisher effect, as found by Westerlund (2008) for a panel of OECD countries. However, if we explicitly introduce the common factors in the Fisher equation, the CupBC and CupFM estimators of the slope parameter on inflation are significantly lower than unity, which implies the existence of a ‘partial’ Fisher effect.

Acknowledgements

This work is an outcome of the research projects 05838/PHCS/07, financed by ‘Programa de Generación de Conocimiento Científico de Excelencia de la Fundación Séneca, Agencia de Ciencia y Tecnología de la Región de Murcia', and ECO2008-06238-C02-01/ECON, funded by the Spanish Ministry of Education and Science. We thank Chihwa Kao, Joakim Westerlund and J. Lluís Carrion-i-Silvestre for providing us the GAUSS routines used in this study.

Notes

¹To test for the presence of cross-sectional dependence in the panel we have applied the Ng (2006) test, which confirms that there is a penetrating dependence across Member States in inflation and interest rates. This justifies the use of the common factors approach (Bai and Ng, 2004) for modelling cross-sectional dependence in the estimation of the Fisher equation.

²We have also applied the Banerjee and Carrion-i-Silvestre (2006) panel cointegration test, obtaining similar results to those derived from the Westerlund and Edgerton (2008) test.

³The DOLS β (with two leads and lags) takes a value of 1.023, almost the same value as that of LSDV β.

⁴To test the sensitivity of the estimation of β to structural breaks, we have obtained the CupBC and CupFM estimators of β after introducing dummies to capture those breaks. Our conclusions remain unaltered.